From 0ede61d8589cc2d93aa78230d74ac58b5b8d0244 Mon Sep 17 00:00:00 2001 From: Christian Brauner Date: Fri, 29 Sep 2023 08:45:59 +0200 Subject: file: convert to SLAB_TYPESAFE_BY_RCU In recent discussions around some performance improvements in the file handling area we discussed switching the file cache to rely on SLAB_TYPESAFE_BY_RCU which allows us to get rid of call_rcu() based freeing for files completely. This is a pretty sensitive change overall but it might actually be worth doing. The main downside is the subtlety. The other one is that we should really wait for Jann's patch to land that enables KASAN to handle SLAB_TYPESAFE_BY_RCU UAFs. Currently it doesn't but a patch for this exists. With SLAB_TYPESAFE_BY_RCU objects may be freed and reused multiple times which requires a few changes. So it isn't sufficient anymore to just acquire a reference to the file in question under rcu using atomic_long_inc_not_zero() since the file might have already been recycled and someone else might have bumped the reference. In other words, callers might see reference count bumps from newer users. For this reason it is necessary to verify that the pointer is the same before and after the reference count increment. This pattern can be seen in get_file_rcu() and __files_get_rcu(). In addition, it isn't possible to access or check fields in struct file without first aqcuiring a reference on it. Not doing that was always very dodgy and it was only usable for non-pointer data in struct file. With SLAB_TYPESAFE_BY_RCU it is necessary that callers first acquire a reference under rcu or they must hold the files_lock of the fdtable. Failing to do either one of this is a bug. Thanks to Jann for pointing out that we need to ensure memory ordering between reallocations and pointer check by ensuring that all subsequent loads have a dependency on the second load in get_file_rcu() and providing a fixup that was folded into this patch. Cc: Jann Horn Suggested-by: Linus Torvalds Signed-off-by: Christian Brauner --- include/linux/fdtable.h | 17 +++-------------- 1 file changed, 3 insertions(+), 14 deletions(-) (limited to 'include/linux/fdtable.h') diff --git a/include/linux/fdtable.h b/include/linux/fdtable.h index e066816f3519..bc4c3287a65e 100644 --- a/include/linux/fdtable.h +++ b/include/linux/fdtable.h @@ -98,20 +98,9 @@ static inline struct file *files_lookup_fd_locked(struct files_struct *files, un return files_lookup_fd_raw(files, fd); } -static inline struct file *files_lookup_fd_rcu(struct files_struct *files, unsigned int fd) -{ - RCU_LOCKDEP_WARN(!rcu_read_lock_held(), - "suspicious rcu_dereference_check() usage"); - return files_lookup_fd_raw(files, fd); -} - -static inline struct file *lookup_fd_rcu(unsigned int fd) -{ - return files_lookup_fd_rcu(current->files, fd); -} - -struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd); -struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *fd); +struct file *lookup_fdget_rcu(unsigned int fd); +struct file *task_lookup_fdget_rcu(struct task_struct *task, unsigned int fd); +struct file *task_lookup_next_fdget_rcu(struct task_struct *task, unsigned int *fd); struct task_struct; -- cgit v1.2.3 From 253ca8678d30bcf94410b54476fc1e0f1627a137 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sun, 26 Nov 2023 12:24:38 -0800 Subject: Improve __fget_files_rcu() code generation (and thus __fget_light()) Commit 0ede61d8589c ("file: convert to SLAB_TYPESAFE_BY_RCU") caused a performance regression as reported by the kernel test robot. The __fget_light() function is one of those critical ones for some loads, and the code generation was unnecessarily impacted. Let's just write that function to better. Reported-by: kernel test robot Cc: Christian Brauner Cc: Jann Horn Cc: Mateusz Guzik Closes: https://lore.kernel.org/oe-lkp/202311201406.2022ca3f-oliver.sang@intel.com Signed-off-by: Linus Torvalds Link: https://lore.kernel.org/r/CAHk-=wiCJtLbFWNURB34b9a_R_unaH3CiMRXfkR0-iihB_z68A@mail.gmail.com Signed-off-by: Christian Brauner --- fs/file.c | 51 ++++++++++++++++++++++++++++++++----------------- include/linux/fdtable.h | 17 +++++++++++------ 2 files changed, 44 insertions(+), 24 deletions(-) (limited to 'include/linux/fdtable.h') diff --git a/fs/file.c b/fs/file.c index 5fb0b146e79e..50df31e104a5 100644 --- a/fs/file.c +++ b/fs/file.c @@ -959,31 +959,45 @@ static inline struct file *__fget_files_rcu(struct files_struct *files, struct file *file; struct fdtable *fdt = rcu_dereference_raw(files->fdt); struct file __rcu **fdentry; + unsigned long nospec_mask; - if (unlikely(fd >= fdt->max_fds)) - return NULL; - - fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds); + /* Mask is a 0 for invalid fd's, ~0 for valid ones */ + nospec_mask = array_index_mask_nospec(fd, fdt->max_fds); /* - * Ok, we have a file pointer. However, because we do - * this all locklessly under RCU, we may be racing with - * that file being closed. - * - * Such a race can take two forms: - * - * (a) the file ref already went down to zero and the - * file hasn't been reused yet or the file count - * isn't zero but the file has already been reused. + * fdentry points to the 'fd' offset, or fdt->fd[0]. + * Loading from fdt->fd[0] is always safe, because the + * array always exists. */ - file = __get_file_rcu(fdentry); + fdentry = fdt->fd + (fd & nospec_mask); + + /* Do the load, then mask any invalid result */ + file = rcu_dereference_raw(*fdentry); + file = (void *)(nospec_mask & (unsigned long)file); if (unlikely(!file)) return NULL; - if (unlikely(IS_ERR(file))) + /* + * Ok, we have a file pointer that was valid at + * some point, but it might have become stale since. + * + * We need to confirm it by incrementing the refcount + * and then check the lookup again. + * + * atomic_long_inc_not_zero() gives us a full memory + * barrier. We only really need an 'acquire' one to + * protect the loads below, but we don't have that. + */ + if (unlikely(!atomic_long_inc_not_zero(&file->f_count))) continue; /* + * Such a race can take two forms: + * + * (a) the file ref already went down to zero and the + * file hasn't been reused yet or the file count + * isn't zero but the file has already been reused. + * * (b) the file table entry has changed under us. * Note that we don't need to re-check the 'fdt->fd' * pointer having changed, because it always goes @@ -991,7 +1005,8 @@ static inline struct file *__fget_files_rcu(struct files_struct *files, * * If so, we need to put our ref and try again. */ - if (unlikely(rcu_dereference_raw(files->fdt) != fdt)) { + if (unlikely(file != rcu_dereference_raw(*fdentry)) || + unlikely(rcu_dereference_raw(files->fdt) != fdt)) { fput(file); continue; } @@ -1128,13 +1143,13 @@ static unsigned long __fget_light(unsigned int fd, fmode_t mask) * atomic_read_acquire() pairs with atomic_dec_and_test() in * put_files_struct(). */ - if (atomic_read_acquire(&files->count) == 1) { + if (likely(atomic_read_acquire(&files->count) == 1)) { file = files_lookup_fd_raw(files, fd); if (!file || unlikely(file->f_mode & mask)) return 0; return (unsigned long)file; } else { - file = __fget(fd, mask); + file = __fget_files(files, fd, mask); if (!file) return 0; return FDPUT_FPUT | (unsigned long)file; diff --git a/include/linux/fdtable.h b/include/linux/fdtable.h index bc4c3287a65e..80bd7789bab1 100644 --- a/include/linux/fdtable.h +++ b/include/linux/fdtable.h @@ -83,12 +83,17 @@ struct dentry; static inline struct file *files_lookup_fd_raw(struct files_struct *files, unsigned int fd) { struct fdtable *fdt = rcu_dereference_raw(files->fdt); - - if (fd < fdt->max_fds) { - fd = array_index_nospec(fd, fdt->max_fds); - return rcu_dereference_raw(fdt->fd[fd]); - } - return NULL; + unsigned long mask = array_index_mask_nospec(fd, fdt->max_fds); + struct file *needs_masking; + + /* + * 'mask' is zero for an out-of-bounds fd, all ones for ok. + * 'fd&mask' is 'fd' for ok, or 0 for out of bounds. + * + * Accessing fdt->fd[0] is ok, but needs masking of the result. + */ + needs_masking = rcu_dereference_raw(fdt->fd[fd&mask]); + return (struct file *)(mask & (unsigned long)needs_masking); } static inline struct file *files_lookup_fd_locked(struct files_struct *files, unsigned int fd) -- cgit v1.2.3 From a88c955fcfb49727d0ed86b47410f6555a8e69e4 Mon Sep 17 00:00:00 2001 From: Christian Brauner Date: Thu, 30 Nov 2023 13:49:07 +0100 Subject: file: s/close_fd_get_file()/file_close_fd()/g That really shouldn't have "get" in there as that implies we're bumping the reference count which we don't do at all. We used to but not anmore. Now we're just closing the fd and pick that file from the fdtable without bumping the reference count. Update the wrong documentation while at it. Link: https://lore.kernel.org/r/20231130-vfs-files-fixes-v1-1-e73ca6f4ea83@kernel.org Reviewed-by: Jan Kara Reviewed-by: Jens Axboe Signed-off-by: Christian Brauner --- drivers/android/binder.c | 2 +- fs/file.c | 14 +++++++++----- fs/open.c | 2 +- include/linux/fdtable.h | 2 +- 4 files changed, 12 insertions(+), 8 deletions(-) (limited to 'include/linux/fdtable.h') diff --git a/drivers/android/binder.c b/drivers/android/binder.c index 92128aae2d06..7658103ba760 100644 --- a/drivers/android/binder.c +++ b/drivers/android/binder.c @@ -1921,7 +1921,7 @@ static void binder_deferred_fd_close(int fd) if (!twcb) return; init_task_work(&twcb->twork, binder_do_fd_close); - twcb->file = close_fd_get_file(fd); + twcb->file = file_close_fd(fd); if (twcb->file) { // pin it until binder_do_fd_close(); see comments there get_file(twcb->file); diff --git a/fs/file.c b/fs/file.c index 50df31e104a5..66f04442a384 100644 --- a/fs/file.c +++ b/fs/file.c @@ -796,7 +796,7 @@ int __close_range(unsigned fd, unsigned max_fd, unsigned int flags) } /* - * See close_fd_get_file() below, this variant assumes current->files->file_lock + * See file_close_fd() below, this variant assumes current->files->file_lock * is held. */ struct file *__close_fd_get_file(unsigned int fd) @@ -804,11 +804,15 @@ struct file *__close_fd_get_file(unsigned int fd) return pick_file(current->files, fd); } -/* - * variant of close_fd that gets a ref on the file for later fput. - * The caller must ensure that filp_close() called on the file. +/** + * file_close_fd - return file associated with fd + * @fd: file descriptor to retrieve file for + * + * Doesn't take a separate reference count. + * + * Returns: The file associated with @fd (NULL if @fd is not open) */ -struct file *close_fd_get_file(unsigned int fd) +struct file *file_close_fd(unsigned int fd) { struct files_struct *files = current->files; struct file *file; diff --git a/fs/open.c b/fs/open.c index 02dc608d40d8..48775329f3f1 100644 --- a/fs/open.c +++ b/fs/open.c @@ -1577,7 +1577,7 @@ SYSCALL_DEFINE1(close, unsigned int, fd) int retval; struct file *file; - file = close_fd_get_file(fd); + file = file_close_fd(fd); if (!file) return -EBADF; diff --git a/include/linux/fdtable.h b/include/linux/fdtable.h index 80bd7789bab1..78c8326d74ae 100644 --- a/include/linux/fdtable.h +++ b/include/linux/fdtable.h @@ -119,7 +119,7 @@ int iterate_fd(struct files_struct *, unsigned, extern int close_fd(unsigned int fd); extern int __close_range(unsigned int fd, unsigned int max_fd, unsigned int flags); -extern struct file *close_fd_get_file(unsigned int fd); +extern struct file *file_close_fd(unsigned int fd); extern int unshare_fd(unsigned long unshare_flags, unsigned int max_fds, struct files_struct **new_fdp); -- cgit v1.2.3